学习笔记 - 贝叶斯

#              特征编号       当前分割值           entropy      
best_split = {"feature_idx":0,"feature_value":0,"entropy":np.inf}

for feaure_idx in range(4):

    # print("当前分割特征为", feaure_idx)

    for value in set(X_train[:, feaure_idx]):  # set 去重减少数据量
        # print("分割值：", value)

        left_idx = X_train[:, feaure_idx] <= value
        right_idx = X_train[:, feaure_idx] > value

        #求信息熵:计算每一类的概率，然后把每一类的概率相乘再相加再取负号

        # 求左边样本个数 这里不能用len 因为right_idx里面是True和False，长度一样
        n_left = sum(left_idx)
        n_right = sum(right_idx)

        # 左
        X_left = X_train[left_idx]
        y_left = y_train[left_idx]

        # 左熵
        """
        print("左侧熵：", y_left)
        temp = {0:0, 1:0, 2:0} # 初始化一个temp 第0类为0个，第1类为0个，第2类为0个
        for ele in y_left: # 遍历，是哪一类哪一类就加1
            temp[ele] += 1
        print(temp)
         
        samples = list(temp.values())
        print(samples)
        
        """
        entropy_left = get_entrepy(y_left)
        # print(entropy_left)

        # print([y_left.tolist().count(idx) for idx in range(3)])

        # 右
        X_right = X_train[right_idx]
        y_right = y_train[right_idx]

        # print("右侧熵", y_right)

        entropy_right = get_entrepy(y_right)
        # print(entropy_right)

        # print("向左走", X_left)
        # print("向右走", X_right)

        # 加权平均
        entropy = n_left / (n_left + n_right) * entropy_left + n_right / (n_left + n_right) * entropy_right
        # print(entropy)
        # 另一种写法
        # print(np.average(a = np.array([entropy_left,entropy_right]),weights = np.array([n_left,n_right])))

        # 判断是否为最佳 - 野蛮计算 穷举思想
        if entropy < best_split["entropy"]:
            best_split["feature_idx"] = feaure_idx
            best_split['feature_value'] = value
            best_split["entropy"] = entropy